The player exploited a weakness in one of the artificial intelligence systems that was discovered by other programs.
A strong amateur Go player has beaten a high-ranking artificial intelligence system after exploiting a weakness discovered by a second computer.
As reported by The Financial Times , by exploiting the flaw, American gamer Kellin Pelrine decisively defeated the KataGo system , winning 14 out of 15 games without further help from the computer.
The victory is one of the greatest progress of humans against the machine after, in 2016, AlphaGo defeated the professionals of this game, marking its leadership.
Pelrine’s victory was made possible by a research company called FAR AI , which developed a program to investigate KataGo ‘s weaknesses .
After playing over a million games, he was able to find a weakness that could be exploited by a decent hobbyist gamer. “It’s not entirely trivial, but it’s not super hard” to learn, Pelrine said. She used the same method to defeat Leela Zero, another of the best AIs in the game.
According to the computer, the best way to beat the AI was to create a large “loop” of stones to surround an opponent’s group, then distract it by making moves in other areas of the board.
Even when his group was almost surrounded, the computer didn’t realize the strategy. “As a human, it would be quite easy to spot,” Pelrine said, as the stones around him stand out clearly on the board.
Failures in the systems
The flaw demonstrates that AI systems can’t really “think” beyond their training.
Lightvector (the developer of KataGo ) is certainly aware of the issue, which players have been exploiting for several months. In a GitHub post, he said that he has been working on a fix for a variety of types of attacks that use the exploit.